Connect your AI assistant

Aztra runs an MCP server — the open Model Context Protocol that AI assistants use to work with external tools. Connect it once and your assistant can read and update issues, sprints, goals, wiki pages and time logs directly, acting as you and limited to what you can already see and do.

The server address is the same for every client:

https://aztra.my/api/mcp/mcp

It is also listed on the Official MCP Registry as my.aztra/aztra, so clients that browse the registry will find it there.

Two ways to connect

Sign in is the better default: nothing to copy, nothing that can leak, and revoking is one click. Use an agent key where a client cannot run an OAuth flow, or where a read-only key is the point.

Claude.ai and Claude Desktop

ChatGPT

Add Aztra as a custom connector using the server address, and sign in when prompted. Custom connectors currently require ChatGPT's developer mode to be enabled for your account or workspace.

Claude Code

Mint a key under Settings → Profile → MCP keys, then run one command:

claude mcp add --transport http aztra "https://aztra.my/api/mcp/mcp" \
  --header "Authorization: Bearer <your key>" \
  --header "Accept: application/json, text/event-stream"

The dialog that shows your new key gives you this command already filled in.

Cursor

Add to .cursor/mcp.json in your project, or the global one in ~/.cursor/:

{
  "mcpServers": {
    "aztra": {
      "url": "https://aztra.my/api/mcp/mcp",
      "headers": { "Authorization": "Bearer <your key>" }
    }
  }
}

VS Code

Add to .vscode/mcp.json:

{
  "servers": {
    "aztra": {
      "type": "http",
      "url": "https://aztra.my/api/mcp/mcp",
      "headers": { "Authorization": "Bearer <your key>" }
    }
  }
}

Any other client

Any client that speaks remote MCP over HTTP takes the same shape — a url and an Authorization header:

{
  "mcpServers": {
    "aztra": {
      "type": "http",
      "url": "https://aztra.my/api/mcp/mcp",
      "headers": {
        "Authorization": "Bearer <your key>",
        "Accept": "application/json, text/event-stream"
      }
    }
  }
}

For clients that only run local (stdio) servers, bridge with mcp-remote:

{
  "mcpServers": {
    "aztra": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://aztra.my/api/mcp/mcp", "--header", "Authorization:Bearer <your key>"]
    }
  }
}

Read-only and read-write keys

When you mint a key you choose its level, and that decides what your assistant is offered:

• Read-only — the assistant sees the read tools only. Write tools are not listed, so it cannot attempt one; there is nothing to refuse.
• Read-write — the assistant sees every tool. Clients that support MCP annotations use them to decide when to ask you first: read tools run without a prompt, and tools marked as changing data ask for confirmation.

A key is bound to one person and inherits their project access. It can never do more than that person can in the app.

What your assistant can do

Every tool the server offers, generated from the same definition the server runs. Names are what the assistant calls; titles are what you will see in its permission prompts.

Read tools (29)

Available to every key. They look things up and never change anything.

Write tools (31)

Offered only to a read-write key. A read-only key never sees them, so an assistant on a read-only key cannot attempt one by mistake.

Limits

• 100 requests a minute per key. Past that the server answers 429 with a Retry-After header, and well-behaved clients wait it out. Normal use is nowhere near this.
• Large results are trimmed. Images embedded in an issue are replaced with a note of their size rather than sent as raw data, and a result too large for the assistant to hold comes back as a short notice with a hint to narrow the request.
• Two writes are gated. Moving an issue to In review or Completed requires time logged on it — the same rule the app enforces.

Revoking access

• Signed-in assistants — Settings → Profile → Connected assistants lists every assistant you have signed in from, with its access level and the workspace it was granted. Disconnect cuts it off immediately: its tokens stop working on the next request, and it has to sign in again to reconnect. Disconnecting inside the client also works, but only Aztra's side is instant.
• Keys — Settings → Profile → MCP keys lists every key with when it was last used and from where. Revoke is instant and permanent; mint a new key if you need one again.